Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ 24641185

History | View | Annotate | Download (12.4 KB)

1
/*
2
 * DSP utils
3
 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
4
 *
5
 * This library is free software; you can redistribute it and/or
6
 * modify it under the terms of the GNU Lesser General Public
7
 * License as published by the Free Software Foundation; either
8
 * version 2 of the License, or (at your option) any later version.
9
 *
10
 * This library is distributed in the hope that it will be useful,
11
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13
 * Lesser General Public License for more details.
14
 *
15
 * You should have received a copy of the GNU Lesser General Public
16
 * License along with this library; if not, write to the Free Software
17
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
18
 */
19

    
20
/**
21
 * @file dsputil.h
22
 * @brief DSP utils
23
 *
24
 */
25

    
26
#ifndef DSPUTIL_H
27
#define DSPUTIL_H
28

    
29
#include "common.h"
30
#include "avcodec.h"
31

    
32

    
33
//#define DEBUG
34
/* dct code */
35
typedef short DCTELEM;
36
//typedef int DCTELEM;
37

    
38
void fdct_ifast (DCTELEM *data);
39
void ff_jpeg_fdct_islow (DCTELEM *data);
40

    
41
void j_rev_dct (DCTELEM *data);
42

    
43
void ff_fdct_mmx(DCTELEM *block);
44

    
45
/* encoding scans */
46
extern const uint8_t ff_alternate_horizontal_scan[64];
47
extern const uint8_t ff_alternate_vertical_scan[64];
48
extern const uint8_t ff_zigzag_direct[64];
49

    
50
/* pixel operations */
51
#define MAX_NEG_CROP 384
52

    
53
/* temporary */
54
extern uint32_t squareTbl[512];
55
extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
56

    
57

    
58
/* minimum alignment rules ;)
59
if u notice errors in the align stuff, need more alignment for some asm code for some cpu
60
or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
61

62
!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
63
i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
64

65
!future video codecs might need functions with less strict alignment
66
*/
67

    
68
/*
69
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
70
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
71
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
72
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
73
void clear_blocks_c(DCTELEM *blocks);
74
*/
75

    
76
/* add and put pixel (decoding) */
77
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
78
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
79
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
80

    
81
#define DEF_OLD_QPEL(name)\
82
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
83
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
84
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
85

    
86
DEF_OLD_QPEL(qpel16_mc11_old_c)
87
DEF_OLD_QPEL(qpel16_mc31_old_c)
88
DEF_OLD_QPEL(qpel16_mc12_old_c)
89
DEF_OLD_QPEL(qpel16_mc32_old_c)
90
DEF_OLD_QPEL(qpel16_mc13_old_c)
91
DEF_OLD_QPEL(qpel16_mc33_old_c)
92
DEF_OLD_QPEL(qpel8_mc11_old_c)
93
DEF_OLD_QPEL(qpel8_mc31_old_c)
94
DEF_OLD_QPEL(qpel8_mc12_old_c)
95
DEF_OLD_QPEL(qpel8_mc32_old_c)
96
DEF_OLD_QPEL(qpel8_mc13_old_c)
97
DEF_OLD_QPEL(qpel8_mc33_old_c)
98

    
99
#define CALL_2X_PIXELS(a, b, n)\
100
static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
101
    b(block  , pixels  , line_size, h);\
102
    b(block+n, pixels+n, line_size, h);\
103
}
104

    
105
/* motion estimation */
106

    
107
typedef int (*op_pixels_abs_func)(uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
108

    
109
typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
110

    
111
/**
112
 * DSPContext.
113
 */
114
typedef struct DSPContext {
115
    /* pixel ops : interface with DCT */
116
    void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
117
    void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
118
    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
119
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
120
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
121
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
122
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
123
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
124
    int (*pix_sum)(uint8_t * pix, int line_size);
125
    int (*pix_norm1)(uint8_t * pix, int line_size);
126
    me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */
127
    me_cmp_func sse[2];
128
    me_cmp_func hadamard8_diff[2];
129
    me_cmp_func dct_sad[2];
130
    me_cmp_func quant_psnr[2];
131
    me_cmp_func bit[2];
132
    me_cmp_func rd[2];
133
    int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
134

    
135
    me_cmp_func me_pre_cmp[11];
136
    me_cmp_func me_cmp[11];
137
    me_cmp_func me_sub_cmp[11];
138
    me_cmp_func mb_cmp[11];
139

    
140
    /* maybe create an array for 16/8 functions */
141
    op_pixels_func put_pixels_tab[2][4];
142
    op_pixels_func avg_pixels_tab[2][4];
143
    op_pixels_func put_no_rnd_pixels_tab[2][4];
144
    op_pixels_func avg_no_rnd_pixels_tab[2][4];
145
    qpel_mc_func put_qpel_pixels_tab[2][16];
146
    qpel_mc_func avg_qpel_pixels_tab[2][16];
147
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
148
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
149
    qpel_mc_func put_mspel_pixels_tab[8];
150

    
151
    op_pixels_abs_func pix_abs16x16;
152
    op_pixels_abs_func pix_abs16x16_x2;
153
    op_pixels_abs_func pix_abs16x16_y2;
154
    op_pixels_abs_func pix_abs16x16_xy2;
155
    op_pixels_abs_func pix_abs8x8;
156
    op_pixels_abs_func pix_abs8x8_x2;
157
    op_pixels_abs_func pix_abs8x8_y2;
158
    op_pixels_abs_func pix_abs8x8_xy2;
159
    
160
    /* huffyuv specific */
161
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
162
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
163
    
164
    /* (I)DCT */
165
    void (*fdct)(DCTELEM *block/* align 16*/);
166
    
167
    /**
168
     * block -> idct -> clip to unsigned 8 bit -> dest.<br>
169
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
170
     * @param line_size size in pixels of a horizotal line of dest
171
     */
172
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
173
    
174
    /**
175
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
176
     * @param line_size size in pixels of a horizotal line of dest
177
     */
178
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
179
    
180
    /**
181
     * idct input permutation.<br>
182
     * an example to avoid confusion:
183
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
184
     * - (x -> referece dct -> reference idct -> x)
185
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
186
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
187
     */
188
    uint8_t idct_permutation[64];
189
    int idct_permutation_type;
190
#define FF_NO_IDCT_PERM 1
191
#define FF_LIBMPEG2_IDCT_PERM 2
192
#define FF_SIMPLE_IDCT_PERM 3
193
#define FF_TRANSPOSE_IDCT_PERM 4
194

    
195
} DSPContext;
196

    
197
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
198

    
199
/**
200
 * permute block according to permuatation.
201
 * @param last last non zero element in scantable order
202
 */
203
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
204

    
205
/**
206
 * Empty mmx state.<br>
207
 * this must be called between any dsp function and float/double code.
208
 * for example sin(); dsp->idct_put(); emms_c(); cos()
209
 */
210
#define emms_c()
211

    
212
/* should be defined by architectures supporting
213
   one or more MultiMedia extension */
214
int mm_support(void);
215

    
216
#if defined(HAVE_MMX)
217

    
218
#undef emms_c
219

    
220
#define MM_MMX    0x0001 /* standard MMX */
221
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
222
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
223
#define MM_SSE    0x0008 /* SSE functions */
224
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
225

    
226
extern int mm_flags;
227

    
228
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
229
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
230

    
231
static inline void emms(void)
232
{
233
    __asm __volatile ("emms;":::"memory");
234
}
235

    
236

    
237
#define emms_c() \
238
{\
239
    if (mm_flags & MM_MMX)\
240
        emms();\
241
}
242

    
243
#define __align8 __attribute__ ((aligned (8)))
244

    
245
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
246
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
247

    
248
#elif defined(ARCH_ARMV4L)
249

    
250
/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
251
   line ptimizations */
252
#define __align8 __attribute__ ((aligned (4)))
253

    
254
void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
255

    
256
#elif defined(HAVE_MLIB)
257

    
258
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
259
#define __align8 __attribute__ ((aligned (8)))
260

    
261
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
262

    
263
#elif defined(ARCH_ALPHA)
264

    
265
#define __align8 __attribute__ ((aligned (8)))
266

    
267
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
268

    
269
#elif defined(ARCH_POWERPC)
270

    
271
#define MM_ALTIVEC    0x0001 /* standard AltiVec */
272

    
273
extern int mm_flags;
274

    
275
#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
276
#include <altivec.h>
277
#endif
278

    
279
#define __align8 __attribute__ ((aligned (16)))
280

    
281
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
282

    
283
#elif defined(HAVE_MMI)
284

    
285
#define __align8 __attribute__ ((aligned (16)))
286

    
287
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
288

    
289
#else
290

    
291
#define __align8
292

    
293
#endif
294

    
295
#ifdef __GNUC__
296

    
297
struct unaligned_64 { uint64_t l; } __attribute__((packed));
298
struct unaligned_32 { uint32_t l; } __attribute__((packed));
299

    
300
#define LD32(a) (((const struct unaligned_32 *) (a))->l)
301
#define LD64(a) (((const struct unaligned_64 *) (a))->l)
302

    
303
#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
304

    
305
#else /* __GNUC__ */
306

    
307
#define LD32(a) (*((uint32_t*)(a)))
308
#define LD64(a) (*((uint64_t*)(a)))
309

    
310
#define ST32(a, b) *((uint32_t*)(a)) = (b)
311

    
312
#endif /* !__GNUC__ */
313

    
314
/* PSNR */
315
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
316
              int orig_linesize[3], int coded_linesize,
317
              AVCodecContext *avctx);
318

    
319
/* FFT computation */
320

    
321
/* NOTE: soon integer code will be added, so you must use the
322
   FFTSample type */
323
typedef float FFTSample;
324

    
325
typedef struct FFTComplex {
326
    FFTSample re, im;
327
} FFTComplex;
328

    
329
typedef struct FFTContext {
330
    int nbits;
331
    int inverse;
332
    uint16_t *revtab;
333
    FFTComplex *exptab;
334
    FFTComplex *exptab1; /* only used by SSE code */
335
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
336
} FFTContext;
337

    
338
int fft_init(FFTContext *s, int nbits, int inverse);
339
void fft_permute(FFTContext *s, FFTComplex *z);
340
void fft_calc_c(FFTContext *s, FFTComplex *z);
341
void fft_calc_sse(FFTContext *s, FFTComplex *z);
342
void fft_calc_altivec(FFTContext *s, FFTComplex *z);
343

    
344
static inline void fft_calc(FFTContext *s, FFTComplex *z)
345
{
346
    s->fft_calc(s, z);
347
}
348
void fft_end(FFTContext *s);
349

    
350
/* MDCT computation */
351

    
352
typedef struct MDCTContext {
353
    int n;  /* size of MDCT (i.e. number of input data * 2) */
354
    int nbits; /* n = 2^nbits */
355
    /* pre/post rotation tables */
356
    FFTSample *tcos;
357
    FFTSample *tsin;
358
    FFTContext fft;
359
} MDCTContext;
360

    
361
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
362
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
363
                const FFTSample *input, FFTSample *tmp);
364
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
365
               const FFTSample *input, FFTSample *tmp);
366
void ff_mdct_end(MDCTContext *s);
367

    
368
#define WARPER88_1616(name8, name16)\
369
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
370
    return name8(s, dst           , src           , stride)\
371
          +name8(s, dst+8         , src+8         , stride)\
372
          +name8(s, dst  +8*stride, src  +8*stride, stride)\
373
          +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
374
}
375

    
376
#ifndef HAVE_LRINTF
377
/* XXX: add ISOC specific test to avoid specific BSD testing. */
378
/* better than nothing implementation. */
379
/* btw, rintf() is existing on fbsd too -- alex */
380
static inline long int lrintf(float x)
381
{
382
#ifdef CONFIG_WIN32
383
    /* XXX: incorrect, but make it compile */
384
    return (int)(x);
385
#else
386
    return (int)(rint(x));
387
#endif
388
}
389
#endif
390

    
391
#endif